Hello from Editor Z. A little over a month after Tesla's Cybercab started carrying paying riders with no safety monitor in the car, the fleet in Austin still can't work a full day. On October 3, 2026, Tesla pushed its Tesla Robotaxi closing time back from 10 p.m. to 11 p.m. — still an hour earlier than the midnight cutoff the service launched with in June 2025. Elon Musk's explanation for why a car that can supposedly drive itself can't be trusted after dark came down to a single image: "grey kittens on grey tarmac in the dark." That one line has reopened a fight Tesla has been having with the rest of the industry since 2019 — whether a camera-only self-driving system can ever really work without lidar — right as Waymo keeps running its own robotaxis around the clock.
A Fleet That's Growing in the Daylight, Shrinking at Night
169 Cars, Home by 11 PM
The numbers on paper look like fast growth. Tesla registered roughly 45 Cybercabs for commercial robotaxi use in Texas when the service opened to paying riders on September 3, 2026. A month later, that number had nearly quadrupled to 169, with 111 vehicles added in a two-week stretch and 43 of those landing on just October 1 and 2 alone, according to Not a Tesla App. Musk has framed the pace as deliberate, saying Tesla is "being extremely careful with autonomous safety, just as we are with making Teslas the safest cars in the world for human driving."
Hours That Have Gone Backward, Not Forward
But fleet size and operating hours are telling two different stories. The service launched in June 2025 running 6 a.m. to midnight. Fifteen months later, after years of promising round-the-clock robotaxis, Tesla is running 6 a.m. to 11 p.m. — an hour shorter than day one, not longer. A bigger fleet sitting idle for more hours a day isn't the kind of scaling investors were expecting when Musk talked up robotaxi economics as potentially worth more than Tesla's entire car business.
Musk's Explanation: Grey Kittens on Grey Tarmac
The Quotes
Asked directly about the night cutoff, Musk didn't point to regulation, staffing, or demand. He said: "The main thing we're trying to solve is making sure that we don't run over pets when they're hard to see at night." Pressed further, he got more specific: "Literally trying to avoid grey kittens on grey tarmac in the dark." It's a vivid image, and it's also a precise technical admission — Tesla is describing a scenario where its cameras can't reliably pick out a small, low-contrast object against a similarly colored background once the ambient light drops.
Why Low Contrast Breaks a Camera-Only System
That's not a software bug Tesla can patch away with a better neural network update; it's a physical limitation of relying on passive cameras that need reflected light to see anything at all. A grey animal on grey pavement at night is close to a worst-case scenario for contrast-based image recognition — there's little light to begin with, and what little there is doesn't distinguish the animal's edges from the road. Tesla's whole self-driving stack, including the Cybercab, is built on cameras and neural networks with no lidar or radar doing the heavy lifting, which is exactly why this specific failure mode exists in the first place.
The Lidar Argument Tesla Has Spent Seven Years Dismissing
From "Fool's Errand" to a Few Hundred Dollars
Musk's skepticism of lidar is not new, which is part of why this moment stings. At Tesla's 2019 Autonomy Day, he called lidar sensors a "fool's errand," and in February 2022 he dismissed them again as a "seductive local maximum" — a technology that looks like a shortcut to a solution but actually traps developers in a dead end. The cost argument has also been quietly collapsing in the years since: Rivian has indicated lidar now costs "a few hundred dollars" per vehicle, a fraction of what the sensors cost when Musk first wrote them off.
What Waymo Runs Instead
Waymo never made the camera-only bet. Its vehicles carry 13 cameras, 4 lidar units, and 6 radars each, and the company operates roughly 4,000 driverless vehicles that run 24 hours a day. Lidar actively emits laser pulses and measures their return, so it works identically in daylight or pitch dark — it doesn't need to distinguish a grey kitten from grey tarmac by contrast, because it's measuring distance and shape directly, not interpreting an image. That's the practical reason Waymo doesn't have an 11 p.m. curfew and Tesla does.
Tesla vs. Waymo: The Night-Driving Gap in Numbers
| Metric | Tesla Robotaxi (Austin) | Waymo |
|---|---|---|
| Sensor approach | Cameras only, no lidar or radar | 13 cameras + 4 lidar + 6 radar per vehicle |
| Operating hours | 6 a.m.–11 p.m. | 24 hours a day |
| Fleet size (as of early Oct. 2026) | 169 Cybercabs in Austin | ~4,000 driverless vehicles across all markets |
| Launch date | June 2025 (driverless rides from Sept. 3, 2026) | Earlier, now in multiple 24/7 metro markets |
| Stated night-driving blocker | Detecting small, low-contrast animals in the dark | None reported — lidar works independent of ambient light |
| Approx. lidar cost per unit | $0 (not used) | "A few hundred dollars" per Rivian's current estimate |
A Rocky First Month Beyond the Dark
Rider Complaints and Navigation Errors
The night curfew isn't the only rough edge CNBC found in Austin's first month. Riders reported long wait times, pickups and drop-offs happening in the wrong spot, and mechanical hiccups with the Cybercab's butterfly doors and trunk not closing properly. The fleet has also struggled navigating Austin's denser downtown corridors and the unpredictable traffic around venues tied to events like South by Southwest — the same kind of early stumbles Waymo and Cruise both went through when they first expanded into dense urban cores.
First Responders Want a Say
Beyond passenger complaints, Austin Fire Captain Matt McElearney has asked for state or national regulation that would let emergency responders move or shut down autonomous vehicles themselves — a car with no steering wheel or pedals can't simply be waved aside or manually pulled over at a fire or crash scene the way a human-driven car can. CNBC noted no major safety incidents have been reported in the fleet's first month, but the first-responder issue, combined with an ongoing federal safety audit, is still unresolved heading into expansion.
What Expansion Actually Requires
Texas, Nevada, Florida — Not Yet "Very, Very Widespread"
Musk's next stated targets are growth within Texas and into Nevada and Florida, with Houston, Dallas, and Miami named as likely candidates for additional Cybercabs. But CNBC points out that the robotaxi network is still nowhere near the "very, very widespread" footprint Musk predicted at Davos back in January 2026. Tesla has reportedly aimed for a multi-city rollout by early 2027, a target that now has to clear both the animal-detection problem and the rider-experience issues from Austin's first month.
The Trade-Off Tesla Keeps Making
None of this means Tesla's camera-only bet is dead — it's cheaper per vehicle to build, and Musk has staked years of engineering and messaging on cameras eventually matching or beating lidar through better neural networks. But "eventually" is doing a lot of work when the fleet in front of riders today still can't drive past 11 p.m. without risking a kitten it may not see coming.
What to Watch Next
The real test for Tesla isn't whether it can register more Cybercabs — 169 in a month shows it can. It's whether those extra hours each week start closing the gap to 24/7, or whether the grey-kitten problem turns out to be a harder wall than Musk's timeline admits. If Waymo keeps running its lidar-equipped fleet through the night without incident while Tesla's curfew holds steady or creeps earlier, that comparison will keep getting harder for Tesla to wave off as a "local maximum."
-EditorZ
Photo by Francisco Gonzalez on Unsplash
Post a Comment